On this page you will find details and instructions on how to set up an environment that supports Intel's AI Pipelines container build and test infrastructure.
Only Linux systems are currently supported. Please make sure the following are installed in your package manager of choice:
makedocker.io
A full installation of docker engine with docker CLI is required. The recommended docker engine version is 19.03.0+.
docker-compose
The Docker Compose CLI can be installed both manually and via package manager.
$ DOCKER_CONFIG=${DOCKER_CONFIG:-$HOME/.docker}
$ mkdir -p $DOCKER_CONFIG/cli-plugins
$ curl -SL https://github.com/docker/compose/releases/download/v2.7.0/docker-compose-linux-x86_64 -o $DOCKER_CONFIG/cli-plugins/docker-compose
$ chmod +x $DOCKER_CONFIG/cli-plugins/docker-compose
$ docker compose version
Docker Compose version v2.7.0
Each pipeline will contain specific requirements and instructions for how to provide its specific dependencies and what customization options are possible. Generally, pipelines are run with the following format:
git submodule update --init --recursive
This will pull the dependent repo containing the scripts to run the end2end pipeline's inference and/or training.
<KEY>=<VALUE> ... <KEY>=<VALUE> make <PIPELINE_NAME>
Where KEY and VALUE pairs are environment variables that can be used to customize both the pipeline's script options and the resulting container. For more information about the valid KEY and VALUE pairs, see the README.md file in the folder for each workflow container:
| AI Workflow | Framework/Tool | Mode |
|---|---|---|
| Document-Level Sentiment Analysis | PyTorch* | Training |
| Vision Based Transfer Learning | TensorFlow | Training | Inference |
| NLP workflow for Azure ML | PyTorch* and Jupyter | Training | Inference |
| Video Streamer | TensorFlow | Inference |
| Quantization Aware Training and Inference | OpenVINO™ | Quantization Aware Training(QAT) |
| Habana® Gaudi® Processor Training and Inference using OpenVINO™ Toolkit for U-Net 2D Model | OpenVINO™ | Training and Inference |
| Wafer Insights | SKLearn | Inference |
Each pipeline can remove all resources allocated by executing make clean.